Needless to say, this is yet another reason why preemptive regulation of AI in the US is very unlikely in the immediate future, for better or worse.
@emollick
-
Public Data Usage in AI Training: Implications and Considerations
By
–
Since this is posted publicly and will be part of future training data, the answer is yes.
-
Runway Aleph video consistency: Woman on mechanical snail pursued by police
By
–
One of the big issues with AI videos is consistency across scenes. It isn't there, but it is getting closer. This is Runway Aleph on a woman riding a snail…
— Ethan Mollick (@emollick) 3 août 2025
"it is night"
"the snail is mechanical"
"show me the front"
"the snail is very fast and is being pursued by police cars" pic.twitter.com/K9l5yXeFI4One of the big issues with AI videos is consistency across scenes. It isn't there, but it is getting closer. This is Runway Aleph on a woman riding a snail…
"it is night"
"the snail is mechanical"
"show me the front"
"the snail is very fast and is being pursued by police cars" -
AI Market Crash Economic Impact Analysis
By
–
A reasonable take on what a crash in the AI market would actually mean for the wider economy, as CapEx for data centers continues to grow. (To be clear, there are no particular warning signs that this is a danger right now, but downside cases are always important to consider).
-
AI Optimization Techniques Show Unpredictable Performance Results
By
–
Sometimes these techniques helped, sometimes they hurt performance. It averaged to almost no effect. There was no clear way to predict in advance which technique would work when. Papers: https://
papers.ssrn.com/sol3/papers.cf
m?abstract_id=5165270
… https://
papers.ssrn.com/sol3/papers.cf
m?abstract_id=5285532
… https://
papers.ssrn.com/sol3/papers.cf
m?abstract_id=5375404
… -
Prompt Engineering Techniques Lose Effectiveness on Recent AI Models
By
–
We have been systematically testing lots of received prompting wisdom & for recent AI models:
Threats, saying please, being insulting, & promising tips do not change average performance on challenging tasks
Chain-of-thought no longer helps even non-reasoner performance much -
AI Acceleration Effects Unknown Within Current Scientific Systems
By
–
To surface a comment from a discussion with @littmath below – as he points out, we don’t know the net effects of this form of acceleration yet This especially true because individual scientists using AI work within systems that may not adapt well to what AI can do.
-
AI Strains Academic Peer Review Systems with Fraud Risks
By
–
Agreed. And our academic & peer review systems are not built for gaining from AI, but instead are designed in a way that is particularly strained by bad AI use (too many papers, worse easy signals of quality, more fraud done better)
-
AI as Research Tool: Time Savings Without Autonomous Breakthroughs
By
–
We do not yet have true autonomous science or breakthrough ideas from AI, but AI provides time savings throughout the process when used carefully by humans: data cleaning, exploratory analysis, writing, pushing back on ideas, (deep) research – all can be helpful when used well.
-
AI Impact in Academia: Time Savings Without Breakthrough Ideas
By
–
Small “a” acceleration so far: no autonomous science, no breakthrough ideas from AI, but time savings throughout the process when used carefully by humans. Academia is very slow. I have a paper that is waiting final acceptance that we started a decade ago.